International Commission for Protection Against Environmental Mutagens and Carcinogens. Application of SAR methods to non-congeneric data bases associated with carcinogenicity and mutagenicity: issues and approaches

Insecticidal and genotoxic potential of two semi-synthetic derivatives of dillapiole for the control of Aedes (Stegomyia) aegypti (Diptera: Culicidae)

The effects of two semi-synthetic dillapiole derivatives, ethyl-ether dillapiole and n-butyl ether dillapiole, on eggs and larvae of Aedes aegypti were studied in view of the need for expansion and renovation of strategic action to control this mosquito - the vector of Dengue virus -, which currently shows a high resistance to chemical insecticides. Eggs and third-instar larvae of A. aegypti that had been exposed to different concentrations of these two compounds showed toxicity and susceptibility, with 100% mortality. Classical cytogenetic assays showed genotoxicity caused by the two compounds in A. aegypti from the cumulative effect of nuclear abnormalities, indicating that these derivatives may be potential alternatives to control A. aegypti.

Expert Systems for Toxicity Prediction

Expert systems offer the facility to predict a toxicity endpoint, as well sometimes as additional relevant information, simply by inputting the chemical structure of a compound. There is now a number of expert systems available, mostly on a commercial basis although a few are free to use or download. This chapter discusses nineteen currently available expert systems, and their performances (if known). Published studies of consensus predictions with these expert systems indicate that these give better results than do individual expert systems.

A test set of compounds with Tetrahymena pyriformis toxicities has been run through the two expert systems known to predict these toxicities; the predictions were quite good, with standard errors of prediction of 0.395 and 0.433 log unit. A further test set of compounds with local lymph node assay skin sensitisation data has been run through seven expert systems, and it was found that consensus predictions were better than were those from any individual expert system.

QSAR modelling of carcinogenicity by balance of correlations

Optimal descriptors based on the simplified molecular input line entry system (SMILES) have been utilized in modeling of carcinogenicity. Carcinogenicity of 401 compounds has been modeled by means of balance of correlations for the training (n = 170) and calibration (n = 170) sets. The obtained models were evaluated with an external test set (n = 61). Comparison of models based on the balance of correlations and models which were obtained on the basis of the total training set (i.e., both training and calibration sets as the united training set) has shown that the balance of correlations improves the statistical quality for the external test set.

Comparative QSTR studies for predicting mutagenicity of nitro compounds

Mutagenicity and carcinogenicity are toxicological endpoints which pose a great concern being the major determinants of cancers and tumours. Nitroarenes possess genotoxic properties as they can form various electrophilic intermediates and adducts with biological systems. Different QSTR techniques were employed to develop models for the prediction of mutagenicity of nitroarenes using a diverse set of 197 nitro aromatic and hetero aromatic molecules. The 2D and 3D QSTR methods used for model development gave statistically significant results. The alignment for 3D methods was obtained by maximum common substructures (MCS) approach, by taking the most mutagenic molecule of the dataset as the template. All the QSTR models were developed with the same set of training and test set molecules. The 3D contours and 2D contribution maps along with molecular fingerprints provide useful information about the mutagenic potentials of the molecules. The GFA based model shows thermodynamic and topological descriptors play an important role in characterizing mutagenicity of nitroarenes. Atomic-level thermodynamic descriptor namely AlogP throws light on hydrophobic features and helps to understand the bilinear model. Topological aspects of these classes of compounds were depicted by the fragment fingerprints and Balaban indices obtained from HQSAR and GFA models, respectively. The predictive abilities of 2D and 3D QSTR models may be useful as a vibrant predictive tool to screen out mutagenic nitroarenes and design safer non-mutagenic nitro compounds.

Identification of the structural requirements for mutagencitiy, by incorporating molecular flexibility and metabolic activation of chemicals. II. General Ames mutagenicity model

The tissue metabolic simulator (TIMES) modeling approach is a hybrid expert system that couples a metabolic simulator together with structure toxicity rules, underpinned by structural alerts, to predict interaction of chemicals or their metabolites with target macromolecules. Some of the structural alerts representing the reactivity pattern-causing effect could interact directly with the target whereas others necessitated a combination with two- or three-dimensional quantitative structure-activity relationship models describing the firing of the alerts from the rest of the molecules. Recently, TIMES has been used to model bacterial mutagenicity [Mekenyan, O., Dimitrov, S., Serafimova, R., Thompson, E., Kotov, S., Dimitrova, N., and Walker, J. (2004) Identification of the structural requirements for mutagenicity by incorporating molecular flexibility and metabolic activation of chemicals I: TA100 model. Chem. Res. Toxicol. 17 (6), 753-766]. The original model was derived for a single tester strain, Salmonella typhimurium (TA100), using the Ames test by the National Toxicology Program (NTP). The model correctly identified 82% of the primary acting mutagens, 94% of the nonmutagens, and 77% of the metabolically activated chemicals in a training set. The identified high correlation between activities across different strains changed the initial strategic direction to look at the other strains in the next modeling developments. In this respect, the focus of the present work was to build a general mutagenicity model predicting mutagenicity with respect to any of the Ames tester strains. The use of all reactivity alerts in the model was justified by their interaction mechanisms with DNA, found in the literature. The alerts identified for the current model were analyzed by comparison with other established alerts derived from human experts. In the new model, the original NTP training set with 1341 structures was expanded by 1626 proprietary chemicals provided by BASF AG. Eventually, the training set consisted of 435 chemicals, which are mutagenic as parents, 397 chemicals that are mutagenic after S9 metabolic activation, and 2012 nonmutagenic chemicals. The general mutagenicity model was found to have 82% sensitivity, 89% specificity, and 88% concordance for training set chemicals. The model applicability domain was introduced accounting for similarity (structural, mechanistic, etc.) between predicted chemicals and training set chemicals for which the model performs correctly.

Assessment of the oral rat chronic lowest observed adverse effect level model in TOPKAT, a QSAR software package for toxicity prediction

The performance of the rat chronic lowest observed adverse effect level (LOAEL, the lowest exposure level at which there are biologically significant increases in the severity of adverse effects) model in Toxicity Prediction by Komputer Assisted Technology (TOPKAT), a commercial quantitative structure-activity relationship software package, was tested on a database of chemicals that are of interest to the U.S. EPA's Office of Pesticide Programs. The testing was repeated on a database of chemicals from three U.S. EPA sources that report peer-reviewed LOAELs. The results of this study were also contrasted with the results of the testing performed during TOPKAT's model-building process.

A topological substructural approach applied to the computational prediction of rodent carcinogenicity

The carcinogenic activity has been investigated by using a topological substructural molecular design approach (TOPS-MODE). A discriminant model was developed to predict the carcinogenic and noncarcinogenic activity on a data set of 189 compounds. The percentage of correct classification was 76.32%. The predictive power of the model was validated by three test: an external test set (compounds not used in the develop of the model, with a 72.97% of good classification), a leave-group-out cross-validation procedure (4-fold full cross-validation, removing 20% of compounds in each cycle, with a good prediction of 76.31%) and two external prediction sets (the first and second exercises of the National Toxicology Program). This methodology evidenced that the hydrophobicity increase the carcinogenic activity and the dipole moment of the molecule decrease it; suggesting the capacity of the TOPS-MODE descriptors to estimate this property for new drug candidates. Finally, the positive and negative fragment contributions to the carcinogenic activity were identified (structural alerts) and their potentialities in the lead generation process and in the design of 'safer' chemicals were evaluated.

Identification of the Structural Requirements for Mutagenicity by Incorporating Molecular Flexibility and Metabolic Activation of Chemicals I: TA100 Model

Traditional attempts to model genotoxicity data have been limited to congeneric data sets, primarily because the mechanism of action was ignored, and frequently, the chemicals required metabolism to the active species. In this exercise, the COmmon REactivity PAtterns (COREPA) approach was used to delineate the structural requirements for eliciting mutagenicity in terms of ranges of descriptors associated with three-dimensional molecular structures. The database used to build the mutagenicity model includes 1196 structurally diverse chemicals tested in the Ames assay by the National Toxicology Program. This manuscript describes the development of the TA100 model that predicts the results of mutagenicity testing using only the Ames TA100 strain. The TA100 model was developed using 148 chemicals that tested positive in TA100 strain without rat liver enzymes (S-9) and 188 chemicals that tested positive in TA100 strain with rat liver enzymes. A decision tree was developed by first comparing the reactivity profile of chemicals that were positive in TA100 without rat liver enzymes to the reactivity profile of the remaining 1048 chemicals. This approach correctly identified 82% of the primary acting mutagens and 94% of the nonmutagens in the training set. The 188 chemicals in the training set that are positive only in the presence of metabolic activation would pass through the decision tree as negative. The next step was to identify the chemicals that are positive only in the presence of metabolic activation. To accomplish this, a series of hierarchically ordered metabolic transformations were used to develop an S-9 metabolism simulator that was applied to each of the 1048 chemicals. The potential metabolites were then screened through the decision tree to identify reactive mutagens. This model correctly identified 77% of the metabolically activated chemicals in a training set. A computer system that applies the COREPA models and predicts mutagenicity of chemicals, including their metabolic activation, was developed. Each prediction is accompanied by a probabilistic estimate of the chemical being in the structural domain covered by the training set.

Chemoinformatics and Drug Discovery

This article reviews current achievements in the field of chemoinformatics and their impact on modern drug discovery processes. The main data mining approaches used in cheminformatics, such as descriptor computations, structural similarity matrices, and classification algorithms, are outlined. The applications of cheminformatics in drug discovery, such as compound selection, virtual library generation, virtual high throughput screening, HTS data mining, and in silico ADMET are discussed. At the conclusion, future directions of chemoinformatics are suggested.

Modelling mutagenicity using properties calculated by computational chemistry

The recent advances in combinatorial chemistry and high throughput screening technologies have led to an explosion in the numbers of possible therapeutic candidates being produced at the early stages of drug discovery. This rapid increase in the number of chemicals to be classified results in a greater need for alternative methods for the prediction of toxicity. Most QSAR models for mutagenicity have been constructed for congeneric series. The prediction requirements of the pharmaceutical industry, however, cover quite diverse chemical structures. This paper reports a study of mutagenicity data for a diverse set of 90 compounds. Good discriminant models have been built for this data set using properties calculated by the techniques of computational chemistry. Jack-knifed (leave one out) predictions for these models are of the order of 85%.

Predicting toxicity: a mechanism of action model of chemical mutagenicity

The increasing importance of theoretical studies for predicting toxicology has aroused the interest of many computational chemists. A new approach has been developed, based on studying at the molecular level two potential mechanisms of action that are related to compound mutagenicity. This approach is the first example that considers both the toxicant and the biological target molecules involved in the interaction. Using some calculated descriptors and a simulation of the interaction chemical, compounds can be classified. More important, the approach helps in understanding and explaining both the correct and the incorrect results, and gives a deeper understanding of the toxic mechanisms. The model has been applied to many compounds and the results are compared with experimental results reported for the corresponding Salmonella tests.

Predicting human safety: screening and computational approaches

Current preclinical safety evaluation programs use a combination of computational methods, mechanistic in vitro screening and - primarily - in vivo experimentation to predict human toxicity. The rapid transition of pharmaceutical R&D into electronic R&D (e-R&D) makes it imperative that predictive safety testing also develops into an information-rich, knowledge-based process in the near future. Accordingly, enhanced databases and computational tools are expected to change the way the pharmaceutical industry assesses drug toxicity during discovery and early development. Expert use of prediction tools should lead to lower failure rates in drug development and decrease the cost and time involved in successful drug approval.

Knowledge discovery and data mining in toxicology

Knowledge discovery and data mining tools are gaining increasing importance for the analysis of toxicological databases. This paper gives a survey of algorithms, capable to derive interpretable models from toxicological data, and presents the most important application areas. The majority of techniques in this area were derived from symbolic machine learning, one commercial product was developed especially for toxicological applications. The main application area is presently the detection of structure-activity relationships, very few authors have used these techniques to solve problems in epidemiological and clinical toxicology. Although the discussed algorithms are very flexible and powerful, further research is required to adopt the algorithms to the specific learning problems in this area, to develop improved representations of chemical and biological data and to enhance the interpretability of the derived models for toxicological experts.

Potential health effects of drinking water disinfection by-products using quantitative structure toxicity relationship

Disinfection by-products (DBPs) are produced as a result of disinfecting water using various treatment methods. Over the years, chlorine has remained the most popular disinfecting agent due to its ability to kill pathogens. However, in 1974, it was discovered that the superchlorination of drinking water resulted in the production of chloroform and other trihalomethanes. Since then hundreds of additional DBPs have been identified, including haloacetic acids and haloacetonitriles with very little or no toxicological data available, thus necessitating the use of additional methods for hazard estimation. Quantitative Structure Toxicity Relationship (QSTR) is one such method and utilizes a computer-based technology to predict the toxicity of a chemical solely from its molecular attributes. The current research was conducted utilizing the TOPKAT/QSTR software package which is comprised of robust, cross-validated QSTR models for assessing mutagenicity, rodent carcinogenicity (female/male; rat/mouse), developmental toxicity, skin sensitization, lowest-observed-adverse-effect level (LOAEL), fathead minnow LC(50), rat oral LD(50) and Daphia magna EC(50). A total of 252 DBPs were analyzed for the likelihood that they would produce tumors and developmental effects using the carcinogenicity and developmental toxicity submodels of TOPKAT. The model predictions were evaluated to identify generalizations between the functional groups (e.g. alcohols, acids, etc.) and specific toxic endpoints. Developmental toxicity was identified as an endpoint common to the majority of aliphatic mono- and dicarboxylic acids, aliphatic halogenated and non-halogenated ketones, and aliphatic haloacetonitriles. In the case of the carcinogenicity submodels, most aliphatic aldehydes were identified as carcinogens only in the female mouse submodel. The majority of the aliphatic and aromatic dicarboxylic acids were identified as carcinogens in the female rat submodel. All other functional groups examined were largely predicted as non-carcinogens in all the cancer submodels (i.e. male/female rats and mice). The QSTR results should aid in the prioritization for evaluation of toxic endpoints in the absence of in vivo bioassays.

Structure-based methods for predicting mutagenicity and carcinogenicity: are we there yet?

There is a great deal of current interest in the use of commercial, automated programs for the prediction of mutagenicity and carcinogenicity based on chemical structure. However, the goal of accurate and reliable toxicity prediction for any chemical, based solely on structural information remains elusive. The toxicity prediction challenge is global in its objective, but limited in its solution, to within local domains of chemicals acting according to similar mechanisms of action in the biological system; to predict, we must be able to generalize based on chemical structure, but the biology fundamentally limits our ability to do so. Available commercial systems for mutagenicity and/or carcinogenicity prediction differ in their specifics, yet most fall in two major categories: (1) automated approaches that rely on the use of statistics for extracting correlations between structure and activity; and (2) knowledge-based expert systems that rely on a set of programmed rules distilled from available knowledge and human expert judgement. These two categories of approaches differ in the ways that they represent, process, and generalize chemical-biological activity information. An application of four commercial systems (TOPKAT, CASE/MULTI-CASE, DEREK, and OncoLogic) to mutagenicity and carcinogenicity prediction for a particular class of chemicals-the haloacetic acids (HAs)-is presented to highlight these differences. Some discussion is devoted to the issue of gauging the relative performance of commercial prediction systems, as well as to the role of prospective prediction exercises in this effort. And finally, an alternative approach that stops short of delivering a prediction to a user, involving structure-searching and data base exploration, is briefly considered.

Quantitative structure-based modeling applied to characterization and prediction of chemical toxicity

Quantitative modeling methods, relating aspects of chemical structure to biological activity, have long been applied to the prediction and characterization of chemical toxicity. The early linear free-energy approaches of Hansch and Free Wilson provided a fundamental scientific framework for the quantitative correlation of chemical structure with biological activity and spurred many developments in the field of quantitative structure-activity relationships (QSARs). In addition to modeling of chemical toxicity, these methods have been extensively applied to modeling of medicinal properties of chemicals. However, there are important differences in the nature and objectives of these two applications, which have led to the evolution of different modeling approaches (namely, the need for treating sets of noncongeneric toxic compounds). In this paper are discussed those approaches to chemical toxicity that have taken a more "personalized" configuration and have undergone implementation into software programs able to perform the various steps of the assessment of the hazard posed by the chemicals. These models focus both on a variety of toxicological endpoints and on key elements of toxicity mechanisms, such as metabolism.

Methods for predicting carcinogenic hazards: new opportunities coming from recent developments in molecular oncology and SAR studies

Without epidemiological evidence, and prior to either short-term tests of genotoxicity or long-term tests of carcinogenicity in rodents, an initial level of information about the carcinogenic hazard of a chemical that perhaps has been designed on paper, but never synthesized, can be provided by structure-activity relationship (SAR) studies. Herein, we have reviewed the interesting strategies developed by human experts and/or computerized approaches for the identification of structural alerts that can denote the possible presence of a carcinogenic hazard in a novel molecule. At a higher level of information, immediately below epidemiological evidence, we have discussed carcinogenicity experiments performed in new types of genetically engineered small rodents. If a dominant oncogene is already mutated, or if an allele of a recessive oncogene is inactivated, we have a model animal with (n-1) stages in the process of carcinogenesis. Both genotoxic and receptor-mediated carcinogens can induce cancers in 20-40% of the time required for classical murine strains. We have described the first interesting results obtained using these new artificial animal models for carcinogenicity studies. We have also briefly discussed other types of engineered mice (lac operon transgenic mice) that are especially suitable for detecting mutagenic effects in a broad spectrum of organs and tissues and that can help to establish mechanistic correlations between mutations and cancer frequencies in specific target organs. Finally, we have reviewed two complementary methods that, while obviously also feasible in rodents, are especially suitable for biomonitoring studies. We have illustrated some of the advantages and drawbacks related to the detection of DNA adducts in target and surrogate tissues using the 32P-DNA postlabeling technique, and we have discussed the possibility of biomonitoring mutations in different human target organs using a molecular technique that combines the activity of restriction enzymes with polymerase chain reaction (RFLP/PCR). Prediction of carcinogenic hazard and biomonitoring are very wide-ranging areas of investigation. We have therefore selected five different subfields for which we felt that interesting innovations have been introduced in the last few years. We have made no attempt to systematically cover the entire area: such an endeavor would have produced a book instead of a review article.

Computational predictive programs (expert systems) in toxicology

The increasing number of pollutants in the environment raises the problem of the toxicological risk evaluation of these chemicals. Several so called expert systems (ES) have been claimed to be able to predict toxicity of certain chemical structures. Different approaches are currently used for these ES, based on explicit rules derived from the knowledge of human experts that compiled lists of toxic moieties for instance in the case of programs called HazardExpert and DEREK or relying on statistical approaches, as in the CASE and TOPKAT programs. Here we describe and compare these and other intelligent computer programs because of their utility in obtaining at least a first rough indication of the potential toxic activity of chemicals.

QSARS of mutagens and carcinogens: two case studies illustrating problems in the construction of models for noncongeneric chemicals

There is a strong motivation to develop QSAR models for toxicity prediction for use in screening, for setting testing priorities, and for reducing reliance on animal testing. Decisions must be made daily by toxicologists in governments and industry to direct limited testing to the most urgent public health problems, and to direct the types of chemical synthesis and product development efforts undertaken. This need has motivated attempts to construct general QSAR models (e.g., for rodent carcinogenicity), not tailored to congeneric series of chemicals. These various attempts have provided interesting and important scientific evidence; however, they have also shared a limited overall performance. The goal of this paper is to illustrate, by two unrelated actual examples of QSARs for mutagens and carcinogens, some fundamental problems relative to the application of general QSAR approaches to noncongeneric chemicals. Both examples consider data sets that are noncongeneric in a chemical structure and mechanism of action sense: in the first case, a mean mutagenic potency defined as an average over multiple genetic toxicity endpoints, and, in the second case, the NTP two-sexes, two species rodent carcinogenicity bioassay results for 280 carcinogens and noncarcinogens. The problems encountered with the QSAR analyses of these two cases indicate that a successful approach to the problem of QSAR modeling of noncongeneric data will need to consider the multidimensional nature of the problem in both a chemical and a biological sense. Since different chemical classes represent largely independent action mechanisms, some means for extracting local QSARs for constituent classes will be necessary. Alternatively, a general QSAR derived for a noncongeneric data set will need to be scrutinized and decomposed along chemical class lines in order to establish boundaries for application and confidence levels for prediction.

The use of computer based structure-activity relationships in the risk assessment of industrial chemicals

The concept of a two-step approach toward the assessment of toxicity endpoints for a chemical is proposed. The first step involves the selection of chemical analogues for which toxicity data is available in a noncongeneric database. The next step is the derivation of a Quantitative Structure-Activity Relationship (QSAR) for the chemical domain, predetermined by the selection rules. The software tools needed for the computer implementation of such an approach are summarized. By making use of them, we have derived aquatic toxicity QSARs, of which two are given as example. The latter pertain to chemicals that have been automatically extracted from noncongeneric databases, after defining the substructure recognition rules implied by the putative mechanism of toxicity.

Alternatives to the 2-species bioassay for the identification of potential human carcinogens

It is proposed that the standard 2-species rodent cancer bioassay protocol, as perfected by the US National Toxicology Program (NTP), has already fulfilled its most useful role by providing an unequalled carcinogenicity database by which to re-assess the type of carcinogen worthy of definition. Continued use of this resource and time consuming protocol can no longer be justified, except in rare circumstances of high and protracted human exposure to a chemical of unknown carcinogenicity. In those rare instances an enlarged bioassay of three or four test species should perhaps be considered, there being nothing fundamental about the rat/mouse combination. In the large majority of cases, however, a practical estimation of the carcinogenic potential of a chemical can be formed in the absence of lifetime carcinogenicity bioassay data. This can be achieved by its sequential study, starting with an appreciation of its chemical structure and anticipated reactivity and mammalian metabolism. After the shortterm evaluation of a range of additional properties of the agent, including its genetic toxicity, rodent toxicity and tissue-specific toxicity, confident predictions of the genotoxic and/or non-genotoxic carcinogenic potential of the agent can be made. In most situations these predictions will be suitable for framing hazard reduction measures among exposed humans. In some situations it may be necessary to evaluate these predicted activities using limited bioassays, a range of which are considered. Extensions of these limited carcinogenicity bioassays to a standard 2-year/2-species bioassay can only be supported in cases where the non-carcinogenicity of the agent becomes the important thing to define. The US NTP have evaluated the carcinogenicity of approximately 400 chemicals over the past 20 years, at a cost of hundreds of millions of US dollars. The experience gained by that and related initiatives, worldwide, can now be harnessed to classify thousands of priority chemicals as being either probable carcinogens or probable noncarcinogens. That can now be achieved using a fraction of the earlier resources and in a fraction of the time that would be required for the conduct of 2-species bioassays. The comfort factor for one group of people of the order of the present system, coupled to the comfort factor for another group of the delay in carcinogenicity assessment enforced by the present council of perfection, are the two main factors delaying transfer to a streamlined system for assessing the carcinogenic potential of chemicals to humans. A third delaying factor in the need for new and focused test data. Coordinated acquisition of such data could rapidly remove the first two obstacles.

Computer prediction of possible toxic action from chemical structure: an update on the DEREK system

Computer-based assessment of potential toxicity has become increasingly popular in recent years. The knowledge-base system DEREK is developed under the guidance of a multinational Collaborative Group of expert toxicologists and provides a qualitative approach to toxicity prediction. Major developments of the DEREK program and knowledge-base have taken place in the last 3 years. Program developments include improvements in both the user interface and data processing. Work on the knowledge-base has concentrated on the areas of genotoxicity and skin sensitisation. DEREK's predictive capabilities for these toxicological end-points has been demonstrated. In addition to the continued expansion of the knowledge-base, a number of enhancements are planned in the DEREK program. In particular, work is in progress to develop further DEREK's ability to report the reasoning behind its predictions.

A computerized connectivity approach for analyzing the structural basis of mutagenicity in Salmonella and its relationship with rodent carcinogenicity

We have applied a new software program, based on graph theory and developed by our group, to predict mutagenicity in Salmonella. The software analyzes, as information in input, the structural formula and the biological activities of a relatively large database of chemicals to generate any possible molecular fragment with size ranging from two to ten nonhydrogen atoms, and detects (as predictors of biological activity) those fragments statistically associated with the biological property investigated. Our previous work used the program to predict carcinogenicity in small rodents. In the current work we applied a modified version of the program, which bases its predictions solely on the most important fragment present in a given molecule, considering as practically negligible the effects of additional less important fragments. For Salmonella mutagenicity we used a database of 551 compounds, and the program achieved a level of predictivity (73.9%) comparable to that obtained by other authors using the Computer Automated Structure Evaluation (CASE) program. We evaluated the relative contributions of biophores and biophobes to overall predictivity: biophores tended to be more important than biophobes, and chemicals containing both biophores and biophobes were more difficult to predict. Many of the molecular fragments identified by the program as being strongly associated with mutagenic activity were similar to the structural alerts identified by the human experts Ashby and Tennant. Our results tend to confirm that structural alerts useful to predict Salmonella mutagenicity are generally not very strong predictors of rodent carcinogenicity. Although the predictivity level achieved for oncogenic activity improved when the program was directly trained with carcinogenicity data, carcinogenicity as a biological endpoint was still more difficult to predict than Salmonella mutagenicity.

Predicting chemical carcinogenesis in rodents: the state of the art in light of a comparative exercise

Within a recent comparative exercise, different approaches to the prediction of rodent carcinogenicity were challenged on a common set of chemicals bioassayed by the U.S. National Toxicology Program. The approaches were of very different natures. Some prediction systems looked for relationships between carcinogenicity and other, more quickly detectable biological events (activity-activity relationships, AAR). Some approaches tended to find structure-activity relationships (SAR). To give an objective evaluation of the results of the exercise, we have analyzed the rodent results and the predictions with the multivariate data analysis methods. The calculated performances varied according to the adopted carcinogenicity classification of the chemicals. When the four rodent results were summarized into a final + or - call, the Tennant approach (AAR method) showed the best performance (about 75% accuracy), whereas the best SAR systems had 60-65% accuracy. A common limitation of almost all the systems was the lack of specificity (too many false positives). Based on these results, better concordance was obtained when the input information was the very costly (and closer to the final endpoint) biological data, rather than the inexpensive (and farther from the endpoint) knowledge of the chemical structure. However, when the rodent results were summarized into a carcinogenicity classification that maintained, to some extent, the gradation intrinsic to the original experimental data, the performance of the AAR systems declined, and the SAR approaches showed a better performance. The difficulty in evaluating the various approaches was further complicated because of a fundamental difference in the approaches themselves: some approaches were 'pure' prediction methods (i.e. their predictions were rigorously based on information not inclusive of carcinogenicity); other approaches (e.g. Tennant, Weisburger) used 'mixed' information, inclusive of known carcinogenicity results from experiments performed before the NTP bioassays. As far as the SAR systems are concerned, their sets of predictions showed a fundamental similarity. This happened in spite of the extremely different procedures adopted to treat the chemical formula (initial information): very simple calculations (Benigni), intuition of the experts (Weisburger and Lijinsky), sophisticated computer programs (TOPKAT and CASE). The results of the Bakale Ke method, based on the experimental measurement of the chemical electrophilicity, and of the Salmonella typhimurium mutagenicity assay were similar to the patterns of predictions of the SAR methods.

A retrospective evaluation of COMPACT predictions of the outcome of NTP rodent carcinogenicity testing

The carcinogenic potentials of 40 National Toxicology Program chemicals previously predicted by Computer Optimised Molecular Parametric Analysis for Chemical Toxicity (COMPACT), based on the identification of potential substrates of cytochromes P4501A and 2E (CYP1A and CYP2E), have been compared with new rodent carcinogenicity results. The COMPACT predictions have also been compared with published Ames mutagenicity data and with our own Hazardexpert predictions for carcinogenicity. Concordance evaluations between rodent carcinogenicity (1/4 segments positive) and predictions by COMPACT or Hazardexpert were 64% for COMPACT (CYP1A only), 72% for COMPACT (CYP1A plus CYP2E), 70% for Hazardexpert alone, and 86% for COMPACT (CYP1A plus CYP2E) plus Hazardexpert. Sensitivities of the predictions were for COMPACT, 75%; Hazardexpert, 60%; and Ames, 54%. Positive predictivities were for COMPACT, 75%; Hazardexpert, 78%; and Ames 81%. Negative predictivites were for COMPACT, 62%; Hazardexpert, 52%; and Ames, 42%.